Deploying TinyML for Energy-Efficient Object Detection and Communication in Low-Power EdgeAI Systems

Abstract The integration of neural networks into low-power controller units is revolutionizing EdgeAI, reducing cloud dependency, latency, and energy consumption while enhancing privacy. This paper presents a compact, energy-efficient system for real-time object detection with TCP/UDP-based image transmission, optimized for low-resource environments. It features an energy-efficient controller, a low-cost camera, and a Wi-Fi module for localized processing and communication. The neural network, trained with deep learning and quantized into TensorFlow Lite, enables high-accuracy object detection on the controller. Real-time images are processed using TinyML, with results transmitted via TCP for reliability or UDP for low-latency tasks. Model compression techniques like quantization, pruning, and hardware-aware deployment optimize performance, while TensorFlow Lite for controllers is integrated. Challenges related to memory, computation, and energy efficiency in constrained environments are addressed. A case study in IoT applications demonstrates the system’s effectiveness in real-time image processing, low-latency transmission, and energy efficiency. The results validate its suitability for smart IoT devices, industrial monitoring, and environmental sensing. This work offers a scalable, cost-effective solution for deploying intelligent systems in remote, resource-limited settings, advancing EdgeAI and IoT technologies.